PostgreSQL Source Code  git master
async.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * async.c
4  * Asynchronous notification: NOTIFY, LISTEN, UNLISTEN
5  *
6  * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  * IDENTIFICATION
10  * src/backend/commands/async.c
11  *
12  *-------------------------------------------------------------------------
13  */
14 
15 /*-------------------------------------------------------------------------
16  * Async Notification Model as of 9.0:
17  *
18  * 1. Multiple backends on same machine. Multiple backends listening on
19  * several channels. (Channels are also called "conditions" in other
20  * parts of the code.)
21  *
22  * 2. There is one central queue in disk-based storage (directory pg_notify/),
23  * with actively-used pages mapped into shared memory by the slru.c module.
24  * All notification messages are placed in the queue and later read out
25  * by listening backends.
26  *
27  * There is no central knowledge of which backend listens on which channel;
28  * every backend has its own list of interesting channels.
29  *
30  * Although there is only one queue, notifications are treated as being
31  * database-local; this is done by including the sender's database OID
32  * in each notification message. Listening backends ignore messages
33  * that don't match their database OID. This is important because it
34  * ensures senders and receivers have the same database encoding and won't
35  * misinterpret non-ASCII text in the channel name or payload string.
36  *
37  * Since notifications are not expected to survive database crashes,
38  * we can simply clean out the pg_notify data at any reboot, and there
39  * is no need for WAL support or fsync'ing.
40  *
41  * 3. Every backend that is listening on at least one channel registers by
42  * entering its PID into the array in AsyncQueueControl. It then scans all
43  * incoming notifications in the central queue and first compares the
44  * database OID of the notification with its own database OID and then
45  * compares the notified channel with the list of channels that it listens
46  * to. In case there is a match it delivers the notification event to its
47  * frontend. Non-matching events are simply skipped.
48  *
49  * 4. The NOTIFY statement (routine Async_Notify) stores the notification in
50  * a backend-local list which will not be processed until transaction end.
51  *
52  * Duplicate notifications from the same transaction are sent out as one
53  * notification only. This is done to save work when for example a trigger
54  * on a 2 million row table fires a notification for each row that has been
55  * changed. If the application needs to receive every single notification
56  * that has been sent, it can easily add some unique string into the extra
57  * payload parameter.
58  *
59  * When the transaction is ready to commit, PreCommit_Notify() adds the
60  * pending notifications to the head of the queue. The head pointer of the
61  * queue always points to the next free position and a position is just a
62  * page number and the offset in that page. This is done before marking the
63  * transaction as committed in clog. If we run into problems writing the
64  * notifications, we can still call elog(ERROR, ...) and the transaction
65  * will roll back.
66  *
67  * Once we have put all of the notifications into the queue, we return to
68  * CommitTransaction() which will then do the actual transaction commit.
69  *
70  * After commit we are called another time (AtCommit_Notify()). Here we
71  * make the actual updates to the effective listen state (listenChannels).
72  *
73  * Finally, after we are out of the transaction altogether, we check if
74  * we need to signal listening backends. In SignalBackends() we scan the
75  * list of listening backends and send a PROCSIG_NOTIFY_INTERRUPT signal
76  * to every listening backend (we don't know which backend is listening on
77  * which channel so we must signal them all). We can exclude backends that
78  * are already up to date, though, and we can also exclude backends that
79  * are in other databases (unless they are way behind and should be kicked
80  * to make them advance their pointers). We don't bother with a
81  * self-signal either, but just process the queue directly.
82  *
83  * 5. Upon receipt of a PROCSIG_NOTIFY_INTERRUPT signal, the signal handler
84  * sets the process's latch, which triggers the event to be processed
85  * immediately if this backend is idle (i.e., it is waiting for a frontend
86  * command and is not within a transaction block. C.f.
87  * ProcessClientReadInterrupt()). Otherwise the handler may only set a
88  * flag, which will cause the processing to occur just before we next go
89  * idle.
90  *
91  * Inbound-notify processing consists of reading all of the notifications
92  * that have arrived since scanning last time. We read every notification
93  * until we reach either a notification from an uncommitted transaction or
94  * the head pointer's position.
95  *
96  * 6. To avoid SLRU wraparound and limit disk space consumption, the tail
97  * pointer needs to be advanced so that old pages can be truncated.
98  * This is relatively expensive (notably, it requires an exclusive lock),
99  * so we don't want to do it often. We make sending backends do this work
100  * if they advanced the queue head into a new page, but only once every
101  * QUEUE_CLEANUP_DELAY pages.
102  *
103  * An application that listens on the same channel it notifies will get
104  * NOTIFY messages for its own NOTIFYs. These can be ignored, if not useful,
105  * by comparing be_pid in the NOTIFY message to the application's own backend's
106  * PID. (As of FE/BE protocol 2.0, the backend's PID is provided to the
107  * frontend during startup.) The above design guarantees that notifies from
108  * other backends will never be missed by ignoring self-notifies.
109  *
110  * The amount of shared memory used for notify management (NUM_NOTIFY_BUFFERS)
111  * can be varied without affecting anything but performance. The maximum
112  * amount of notification data that can be queued at one time is determined
113  * by slru.c's wraparound limit; see QUEUE_MAX_PAGE below.
114  *-------------------------------------------------------------------------
115  */
116 
117 #include "postgres.h"
118 
119 #include <limits.h>
120 #include <unistd.h>
121 #include <signal.h>
122 
123 #include "access/parallel.h"
124 #include "access/slru.h"
125 #include "access/transam.h"
126 #include "access/xact.h"
127 #include "catalog/pg_database.h"
128 #include "commands/async.h"
129 #include "common/hashfn.h"
130 #include "funcapi.h"
131 #include "libpq/libpq.h"
132 #include "libpq/pqformat.h"
133 #include "miscadmin.h"
134 #include "storage/ipc.h"
135 #include "storage/lmgr.h"
136 #include "storage/proc.h"
137 #include "storage/procarray.h"
138 #include "storage/procsignal.h"
139 #include "storage/sinval.h"
140 #include "tcop/tcopprot.h"
141 #include "utils/builtins.h"
142 #include "utils/memutils.h"
143 #include "utils/ps_status.h"
144 #include "utils/snapmgr.h"
145 #include "utils/timestamp.h"
146 
147 
148 /*
149  * Maximum size of a NOTIFY payload, including terminating NULL. This
150  * must be kept small enough so that a notification message fits on one
151  * SLRU page. The magic fudge factor here is noncritical as long as it's
152  * more than AsyncQueueEntryEmptySize --- we make it significantly bigger
153  * than that, so changes in that data structure won't affect user-visible
154  * restrictions.
155  */
156 #define NOTIFY_PAYLOAD_MAX_LENGTH (BLCKSZ - NAMEDATALEN - 128)
157 
158 /*
159  * Struct representing an entry in the global notify queue
160  *
161  * This struct declaration has the maximal length, but in a real queue entry
162  * the data area is only big enough for the actual channel and payload strings
163  * (each null-terminated). AsyncQueueEntryEmptySize is the minimum possible
164  * entry size, if both channel and payload strings are empty (but note it
165  * doesn't include alignment padding).
166  *
167  * The "length" field should always be rounded up to the next QUEUEALIGN
168  * multiple so that all fields are properly aligned.
169  */
170 typedef struct AsyncQueueEntry
171 {
172  int length; /* total allocated length of entry */
173  Oid dboid; /* sender's database OID */
174  TransactionId xid; /* sender's XID */
175  int32 srcPid; /* sender's PID */
178 
179 /* Currently, no field of AsyncQueueEntry requires more than int alignment */
180 #define QUEUEALIGN(len) INTALIGN(len)
181 
182 #define AsyncQueueEntryEmptySize (offsetof(AsyncQueueEntry, data) + 2)
183 
184 /*
185  * Struct describing a queue position, and assorted macros for working with it
186  */
187 typedef struct QueuePosition
188 {
189  int page; /* SLRU page number */
190  int offset; /* byte offset within page */
191 } QueuePosition;
192 
193 #define QUEUE_POS_PAGE(x) ((x).page)
194 #define QUEUE_POS_OFFSET(x) ((x).offset)
195 
196 #define SET_QUEUE_POS(x,y,z) \
197  do { \
198  (x).page = (y); \
199  (x).offset = (z); \
200  } while (0)
201 
202 #define QUEUE_POS_EQUAL(x,y) \
203  ((x).page == (y).page && (x).offset == (y).offset)
204 
205 #define QUEUE_POS_IS_ZERO(x) \
206  ((x).page == 0 && (x).offset == 0)
207 
208 /* choose logically smaller QueuePosition */
209 #define QUEUE_POS_MIN(x,y) \
210  (asyncQueuePagePrecedes((x).page, (y).page) ? (x) : \
211  (x).page != (y).page ? (y) : \
212  (x).offset < (y).offset ? (x) : (y))
213 
214 /* choose logically larger QueuePosition */
215 #define QUEUE_POS_MAX(x,y) \
216  (asyncQueuePagePrecedes((x).page, (y).page) ? (y) : \
217  (x).page != (y).page ? (x) : \
218  (x).offset > (y).offset ? (x) : (y))
219 
220 /*
221  * Parameter determining how often we try to advance the tail pointer:
222  * we do that after every QUEUE_CLEANUP_DELAY pages of NOTIFY data. This is
223  * also the distance by which a backend in another database needs to be
224  * behind before we'll decide we need to wake it up to advance its pointer.
225  *
226  * Resist the temptation to make this really large. While that would save
227  * work in some places, it would add cost in others. In particular, this
228  * should likely be less than NUM_NOTIFY_BUFFERS, to ensure that backends
229  * catch up before the pages they'll need to read fall out of SLRU cache.
230  */
231 #define QUEUE_CLEANUP_DELAY 4
232 
233 /*
234  * Struct describing a listening backend's status
235  */
236 typedef struct QueueBackendStatus
237 {
238  int32 pid; /* either a PID or InvalidPid */
239  Oid dboid; /* backend's database OID, or InvalidOid */
240  BackendId nextListener; /* id of next listener, or InvalidBackendId */
241  QueuePosition pos; /* backend has read queue up to here */
243 
244 /*
245  * Shared memory state for LISTEN/NOTIFY (excluding its SLRU stuff)
246  *
247  * The AsyncQueueControl structure is protected by the NotifyQueueLock and
248  * NotifyQueueTailLock.
249  *
250  * When holding NotifyQueueLock in SHARED mode, backends may only inspect
251  * their own entries as well as the head and tail pointers. Consequently we
252  * can allow a backend to update its own record while holding only SHARED lock
253  * (since no other backend will inspect it).
254  *
255  * When holding NotifyQueueLock in EXCLUSIVE mode, backends can inspect the
256  * entries of other backends and also change the head pointer. When holding
257  * both NotifyQueueLock and NotifyQueueTailLock in EXCLUSIVE mode, backends
258  * can change the tail pointer.
259  *
260  * NotifySLRULock is used as the control lock for the pg_notify SLRU buffers.
261  * In order to avoid deadlocks, whenever we need multiple locks, we first get
262  * NotifyQueueTailLock, then NotifyQueueLock, and lastly NotifySLRULock.
263  *
264  * Each backend uses the backend[] array entry with index equal to its
265  * BackendId (which can range from 1 to MaxBackends). We rely on this to make
266  * SendProcSignal fast.
267  *
268  * The backend[] array entries for actively-listening backends are threaded
269  * together using firstListener and the nextListener links, so that we can
270  * scan them without having to iterate over inactive entries. We keep this
271  * list in order by BackendId so that the scan is cache-friendly when there
272  * are many active entries.
273  */
274 typedef struct AsyncQueueControl
275 {
276  QueuePosition head; /* head points to the next free location */
277  QueuePosition tail; /* tail must be <= the queue position of every
278  * listening backend */
279  BackendId firstListener; /* id of first listener, or InvalidBackendId */
280  TimestampTz lastQueueFillWarn; /* time of last queue-full msg */
282  /* backend[0] is not used; used entries are from [1] to [MaxBackends] */
284 
286 
287 #define QUEUE_HEAD (asyncQueueControl->head)
288 #define QUEUE_TAIL (asyncQueueControl->tail)
289 #define QUEUE_FIRST_LISTENER (asyncQueueControl->firstListener)
290 #define QUEUE_BACKEND_PID(i) (asyncQueueControl->backend[i].pid)
291 #define QUEUE_BACKEND_DBOID(i) (asyncQueueControl->backend[i].dboid)
292 #define QUEUE_NEXT_LISTENER(i) (asyncQueueControl->backend[i].nextListener)
293 #define QUEUE_BACKEND_POS(i) (asyncQueueControl->backend[i].pos)
294 
295 /*
296  * The SLRU buffer area through which we access the notification queue
297  */
299 
300 #define NotifyCtl (&NotifyCtlData)
301 #define QUEUE_PAGESIZE BLCKSZ
302 #define QUEUE_FULL_WARN_INTERVAL 5000 /* warn at most once every 5s */
303 
304 /*
305  * Use segments 0000 through FFFF. Each contains SLRU_PAGES_PER_SEGMENT pages
306  * which gives us the pages from 0 to SLRU_PAGES_PER_SEGMENT * 0x10000 - 1.
307  * We could use as many segments as SlruScanDirectory() allows, but this gives
308  * us so much space already that it doesn't seem worth the trouble.
309  *
310  * The most data we can have in the queue at a time is QUEUE_MAX_PAGE/2
311  * pages, because more than that would confuse slru.c into thinking there
312  * was a wraparound condition. With the default BLCKSZ this means there
313  * can be up to 8GB of queued-and-not-read data.
314  *
315  * Note: it's possible to redefine QUEUE_MAX_PAGE with a smaller multiple of
316  * SLRU_PAGES_PER_SEGMENT, for easier testing of queue-full behaviour.
317  */
318 #define QUEUE_MAX_PAGE (SLRU_PAGES_PER_SEGMENT * 0x10000 - 1)
319 
320 /*
321  * listenChannels identifies the channels we are actually listening to
322  * (ie, have committed a LISTEN on). It is a simple list of channel names,
323  * allocated in TopMemoryContext.
324  */
325 static List *listenChannels = NIL; /* list of C strings */
326 
327 /*
328  * State for pending LISTEN/UNLISTEN actions consists of an ordered list of
329  * all actions requested in the current transaction. As explained above,
330  * we don't actually change listenChannels until we reach transaction commit.
331  *
332  * The list is kept in CurTransactionContext. In subtransactions, each
333  * subtransaction has its own list in its own CurTransactionContext, but
334  * successful subtransactions attach their lists to their parent's list.
335  * Failed subtransactions simply discard their lists.
336  */
337 typedef enum
338 {
343 
344 typedef struct
345 {
347  char channel[FLEXIBLE_ARRAY_MEMBER]; /* nul-terminated string */
348 } ListenAction;
349 
350 typedef struct ActionList
351 {
352  int nestingLevel; /* current transaction nesting depth */
353  List *actions; /* list of ListenAction structs */
354  struct ActionList *upper; /* details for upper transaction levels */
355 } ActionList;
356 
357 static ActionList *pendingActions = NULL;
358 
359 /*
360  * State for outbound notifies consists of a list of all channels+payloads
361  * NOTIFYed in the current transaction. We do not actually perform a NOTIFY
362  * until and unless the transaction commits. pendingNotifies is NULL if no
363  * NOTIFYs have been done in the current (sub) transaction.
364  *
365  * We discard duplicate notify events issued in the same transaction.
366  * Hence, in addition to the list proper (which we need to track the order
367  * of the events, since we guarantee to deliver them in order), we build a
368  * hash table which we can probe to detect duplicates. Since building the
369  * hash table is somewhat expensive, we do so only once we have at least
370  * MIN_HASHABLE_NOTIFIES events queued in the current (sub) transaction;
371  * before that we just scan the events linearly.
372  *
373  * The list is kept in CurTransactionContext. In subtransactions, each
374  * subtransaction has its own list in its own CurTransactionContext, but
375  * successful subtransactions add their entries to their parent's list.
376  * Failed subtransactions simply discard their lists. Since these lists
377  * are independent, there may be notify events in a subtransaction's list
378  * that duplicate events in some ancestor (sub) transaction; we get rid of
379  * the dups when merging the subtransaction's list into its parent's.
380  *
381  * Note: the action and notify lists do not interact within a transaction.
382  * In particular, if a transaction does NOTIFY and then LISTEN on the same
383  * condition name, it will get a self-notify at commit. This is a bit odd
384  * but is consistent with our historical behavior.
385  */
386 typedef struct Notification
387 {
388  uint16 channel_len; /* length of channel-name string */
389  uint16 payload_len; /* length of payload string */
390  /* null-terminated channel name, then null-terminated payload follow */
392 } Notification;
393 
394 typedef struct NotificationList
395 {
396  int nestingLevel; /* current transaction nesting depth */
397  List *events; /* list of Notification structs */
398  HTAB *hashtab; /* hash of NotificationHash structs, or NULL */
399  struct NotificationList *upper; /* details for upper transaction levels */
401 
402 #define MIN_HASHABLE_NOTIFIES 16 /* threshold to build hashtab */
403 
404 typedef struct NotificationHash
405 {
406  Notification *event; /* => the actual Notification struct */
408 
410 
411 /*
412  * Inbound notifications are initially processed by HandleNotifyInterrupt(),
413  * called from inside a signal handler. That just sets the
414  * notifyInterruptPending flag and sets the process
415  * latch. ProcessNotifyInterrupt() will then be called whenever it's safe to
416  * actually deal with the interrupt.
417  */
418 volatile sig_atomic_t notifyInterruptPending = false;
419 
420 /* True if we've registered an on_shmem_exit cleanup */
421 static bool unlistenExitRegistered = false;
422 
423 /* True if we're currently registered as a listener in asyncQueueControl */
424 static bool amRegisteredListener = false;
425 
426 /* has this backend sent notifications in the current transaction? */
427 static bool backendHasSentNotifications = false;
428 
429 /* have we advanced to a page that's a multiple of QUEUE_CLEANUP_DELAY? */
430 static bool backendTryAdvanceTail = false;
431 
432 /* GUC parameter */
433 bool Trace_notify = false;
434 
435 /* local function prototypes */
436 static int asyncQueuePageDiff(int p, int q);
437 static bool asyncQueuePagePrecedes(int p, int q);
438 static void queue_listen(ListenActionKind action, const char *channel);
439 static void Async_UnlistenOnExit(int code, Datum arg);
440 static void Exec_ListenPreCommit(void);
441 static void Exec_ListenCommit(const char *channel);
442 static void Exec_UnlistenCommit(const char *channel);
443 static void Exec_UnlistenAllCommit(void);
444 static bool IsListeningOn(const char *channel);
445 static void asyncQueueUnregister(void);
446 static bool asyncQueueIsFull(void);
447 static bool asyncQueueAdvance(volatile QueuePosition *position, int entryLength);
449 static ListCell *asyncQueueAddEntries(ListCell *nextNotify);
450 static double asyncQueueUsage(void);
451 static void asyncQueueFillWarning(void);
452 static void SignalBackends(void);
453 static void asyncQueueReadAllNotifications(void);
454 static bool asyncQueueProcessPageEntries(volatile QueuePosition *current,
455  QueuePosition stop,
456  char *page_buffer,
457  Snapshot snapshot);
458 static void asyncQueueAdvanceTail(void);
459 static void ProcessIncomingNotify(void);
460 static bool AsyncExistsPendingNotify(Notification *n);
462 static uint32 notification_hash(const void *key, Size keysize);
463 static int notification_match(const void *key1, const void *key2, Size keysize);
464 static void ClearPendingActionsAndNotifies(void);
465 
466 /*
467  * Compute the difference between two queue page numbers (i.e., p - q),
468  * accounting for wraparound.
469  */
470 static int
471 asyncQueuePageDiff(int p, int q)
472 {
473  int diff;
474 
475  /*
476  * We have to compare modulo (QUEUE_MAX_PAGE+1)/2. Both inputs should be
477  * in the range 0..QUEUE_MAX_PAGE.
478  */
479  Assert(p >= 0 && p <= QUEUE_MAX_PAGE);
480  Assert(q >= 0 && q <= QUEUE_MAX_PAGE);
481 
482  diff = p - q;
483  if (diff >= ((QUEUE_MAX_PAGE + 1) / 2))
484  diff -= QUEUE_MAX_PAGE + 1;
485  else if (diff < -((QUEUE_MAX_PAGE + 1) / 2))
486  diff += QUEUE_MAX_PAGE + 1;
487  return diff;
488 }
489 
490 /* Is p < q, accounting for wraparound? */
491 static bool
493 {
494  return asyncQueuePageDiff(p, q) < 0;
495 }
496 
497 /*
498  * Report space needed for our shared memory area
499  */
500 Size
502 {
503  Size size;
504 
505  /* This had better match AsyncShmemInit */
506  size = mul_size(MaxBackends + 1, sizeof(QueueBackendStatus));
507  size = add_size(size, offsetof(AsyncQueueControl, backend));
508 
510 
511  return size;
512 }
513 
514 /*
515  * Initialize our shared memory area
516  */
517 void
519 {
520  bool found;
521  Size size;
522 
523  /*
524  * Create or attach to the AsyncQueueControl structure.
525  *
526  * The used entries in the backend[] array run from 1 to MaxBackends; the
527  * zero'th entry is unused but must be allocated.
528  */
529  size = mul_size(MaxBackends + 1, sizeof(QueueBackendStatus));
530  size = add_size(size, offsetof(AsyncQueueControl, backend));
531 
532  asyncQueueControl = (AsyncQueueControl *)
533  ShmemInitStruct("Async Queue Control", size, &found);
534 
535  if (!found)
536  {
537  /* First time through, so initialize it */
538  SET_QUEUE_POS(QUEUE_HEAD, 0, 0);
539  SET_QUEUE_POS(QUEUE_TAIL, 0, 0);
541  asyncQueueControl->lastQueueFillWarn = 0;
542  /* zero'th entry won't be used, but let's initialize it anyway */
543  for (int i = 0; i <= MaxBackends; i++)
544  {
549  }
550  }
551 
552  /*
553  * Set up SLRU management of the pg_notify data.
554  */
555  NotifyCtl->PagePrecedes = asyncQueuePagePrecedes;
557  NotifySLRULock, "pg_notify", LWTRANCHE_NOTIFY_BUFFER,
559 
560  if (!found)
561  {
562  /*
563  * During start or reboot, clean out the pg_notify directory.
564  */
566  }
567 }
568 
569 
570 /*
571  * pg_notify -
572  * SQL function to send a notification event
573  */
574 Datum
576 {
577  const char *channel;
578  const char *payload;
579 
580  if (PG_ARGISNULL(0))
581  channel = "";
582  else
583  channel = text_to_cstring(PG_GETARG_TEXT_PP(0));
584 
585  if (PG_ARGISNULL(1))
586  payload = "";
587  else
588  payload = text_to_cstring(PG_GETARG_TEXT_PP(1));
589 
590  /* For NOTIFY as a statement, this is checked in ProcessUtility */
592 
593  Async_Notify(channel, payload);
594 
595  PG_RETURN_VOID();
596 }
597 
598 
599 /*
600  * Async_Notify
601  *
602  * This is executed by the SQL notify command.
603  *
604  * Adds the message to the list of pending notifies.
605  * Actual notification happens during transaction commit.
606  * ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
607  */
608 void
609 Async_Notify(const char *channel, const char *payload)
610 {
611  int my_level = GetCurrentTransactionNestLevel();
612  size_t channel_len;
613  size_t payload_len;
614  Notification *n;
615  MemoryContext oldcontext;
616 
617  if (IsParallelWorker())
618  elog(ERROR, "cannot send notifications from a parallel worker");
619 
620  if (Trace_notify)
621  elog(DEBUG1, "Async_Notify(%s)", channel);
622 
623  channel_len = channel ? strlen(channel) : 0;
624  payload_len = payload ? strlen(payload) : 0;
625 
626  /* a channel name must be specified */
627  if (channel_len == 0)
628  ereport(ERROR,
629  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
630  errmsg("channel name cannot be empty")));
631 
632  /* enforce length limits */
633  if (channel_len >= NAMEDATALEN)
634  ereport(ERROR,
635  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
636  errmsg("channel name too long")));
637 
638  if (payload_len >= NOTIFY_PAYLOAD_MAX_LENGTH)
639  ereport(ERROR,
640  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
641  errmsg("payload string too long")));
642 
643  /*
644  * We must construct the Notification entry, even if we end up not using
645  * it, in order to compare it cheaply to existing list entries.
646  *
647  * The notification list needs to live until end of transaction, so store
648  * it in the transaction context.
649  */
651 
653  channel_len + payload_len + 2);
654  n->channel_len = channel_len;
655  n->payload_len = payload_len;
656  strcpy(n->data, channel);
657  if (payload)
658  strcpy(n->data + channel_len + 1, payload);
659  else
660  n->data[channel_len + 1] = '\0';
661 
662  if (pendingNotifies == NULL || my_level > pendingNotifies->nestingLevel)
663  {
664  NotificationList *notifies;
665 
666  /*
667  * First notify event in current (sub)xact. Note that we allocate the
668  * NotificationList in TopTransactionContext; the nestingLevel might
669  * get changed later by AtSubCommit_Notify.
670  */
671  notifies = (NotificationList *)
673  sizeof(NotificationList));
674  notifies->nestingLevel = my_level;
675  notifies->events = list_make1(n);
676  /* We certainly don't need a hashtable yet */
677  notifies->hashtab = NULL;
678  notifies->upper = pendingNotifies;
679  pendingNotifies = notifies;
680  }
681  else
682  {
683  /* Now check for duplicates */
685  {
686  /* It's a dup, so forget it */
687  pfree(n);
688  MemoryContextSwitchTo(oldcontext);
689  return;
690  }
691 
692  /* Append more events to existing list */
694  }
695 
696  MemoryContextSwitchTo(oldcontext);
697 }
698 
699 /*
700  * queue_listen
701  * Common code for listen, unlisten, unlisten all commands.
702  *
703  * Adds the request to the list of pending actions.
704  * Actual update of the listenChannels list happens during transaction
705  * commit.
706  */
707 static void
708 queue_listen(ListenActionKind action, const char *channel)
709 {
710  MemoryContext oldcontext;
711  ListenAction *actrec;
712  int my_level = GetCurrentTransactionNestLevel();
713 
714  /*
715  * Unlike Async_Notify, we don't try to collapse out duplicates. It would
716  * be too complicated to ensure we get the right interactions of
717  * conflicting LISTEN/UNLISTEN/UNLISTEN_ALL, and it's unlikely that there
718  * would be any performance benefit anyway in sane applications.
719  */
721 
722  /* space for terminating null is included in sizeof(ListenAction) */
723  actrec = (ListenAction *) palloc(offsetof(ListenAction, channel) +
724  strlen(channel) + 1);
725  actrec->action = action;
726  strcpy(actrec->channel, channel);
727 
728  if (pendingActions == NULL || my_level > pendingActions->nestingLevel)
729  {
730  ActionList *actions;
731 
732  /*
733  * First action in current sub(xact). Note that we allocate the
734  * ActionList in TopTransactionContext; the nestingLevel might get
735  * changed later by AtSubCommit_Notify.
736  */
737  actions = (ActionList *)
739  actions->nestingLevel = my_level;
740  actions->actions = list_make1(actrec);
741  actions->upper = pendingActions;
742  pendingActions = actions;
743  }
744  else
745  pendingActions->actions = lappend(pendingActions->actions, actrec);
746 
747  MemoryContextSwitchTo(oldcontext);
748 }
749 
750 /*
751  * Async_Listen
752  *
753  * This is executed by the SQL listen command.
754  */
755 void
756 Async_Listen(const char *channel)
757 {
758  if (Trace_notify)
759  elog(DEBUG1, "Async_Listen(%s,%d)", channel, MyProcPid);
760 
761  queue_listen(LISTEN_LISTEN, channel);
762 }
763 
764 /*
765  * Async_Unlisten
766  *
767  * This is executed by the SQL unlisten command.
768  */
769 void
770 Async_Unlisten(const char *channel)
771 {
772  if (Trace_notify)
773  elog(DEBUG1, "Async_Unlisten(%s,%d)", channel, MyProcPid);
774 
775  /* If we couldn't possibly be listening, no need to queue anything */
776  if (pendingActions == NULL && !unlistenExitRegistered)
777  return;
778 
779  queue_listen(LISTEN_UNLISTEN, channel);
780 }
781 
782 /*
783  * Async_UnlistenAll
784  *
785  * This is invoked by UNLISTEN * command, and also at backend exit.
786  */
787 void
789 {
790  if (Trace_notify)
791  elog(DEBUG1, "Async_UnlistenAll(%d)", MyProcPid);
792 
793  /* If we couldn't possibly be listening, no need to queue anything */
794  if (pendingActions == NULL && !unlistenExitRegistered)
795  return;
796 
798 }
799 
800 /*
801  * SQL function: return a set of the channel names this backend is actively
802  * listening to.
803  *
804  * Note: this coding relies on the fact that the listenChannels list cannot
805  * change within a transaction.
806  */
807 Datum
809 {
810  FuncCallContext *funcctx;
811 
812  /* stuff done only on the first call of the function */
813  if (SRF_IS_FIRSTCALL())
814  {
815  /* create a function context for cross-call persistence */
816  funcctx = SRF_FIRSTCALL_INIT();
817  }
818 
819  /* stuff done on every call of the function */
820  funcctx = SRF_PERCALL_SETUP();
821 
822  if (funcctx->call_cntr < list_length(listenChannels))
823  {
824  char *channel = (char *) list_nth(listenChannels,
825  funcctx->call_cntr);
826 
827  SRF_RETURN_NEXT(funcctx, CStringGetTextDatum(channel));
828  }
829 
830  SRF_RETURN_DONE(funcctx);
831 }
832 
833 /*
834  * Async_UnlistenOnExit
835  *
836  * This is executed at backend exit if we have done any LISTENs in this
837  * backend. It might not be necessary anymore, if the user UNLISTENed
838  * everything, but we don't try to detect that case.
839  */
840 static void
842 {
845 }
846 
847 /*
848  * AtPrepare_Notify
849  *
850  * This is called at the prepare phase of a two-phase
851  * transaction. Save the state for possible commit later.
852  */
853 void
855 {
856  /* It's not allowed to have any pending LISTEN/UNLISTEN/NOTIFY actions */
857  if (pendingActions || pendingNotifies)
858  ereport(ERROR,
859  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
860  errmsg("cannot PREPARE a transaction that has executed LISTEN, UNLISTEN, or NOTIFY")));
861 }
862 
863 /*
864  * PreCommit_Notify
865  *
866  * This is called at transaction commit, before actually committing to
867  * clog.
868  *
869  * If there are pending LISTEN actions, make sure we are listed in the
870  * shared-memory listener array. This must happen before commit to
871  * ensure we don't miss any notifies from transactions that commit
872  * just after ours.
873  *
874  * If there are outbound notify requests in the pendingNotifies list,
875  * add them to the global queue. We do that before commit so that
876  * we can still throw error if we run out of queue space.
877  */
878 void
880 {
881  ListCell *p;
882 
883  if (!pendingActions && !pendingNotifies)
884  return; /* no relevant statements in this xact */
885 
886  if (Trace_notify)
887  elog(DEBUG1, "PreCommit_Notify");
888 
889  /* Preflight for any pending listen/unlisten actions */
890  if (pendingActions != NULL)
891  {
892  foreach(p, pendingActions->actions)
893  {
894  ListenAction *actrec = (ListenAction *) lfirst(p);
895 
896  switch (actrec->action)
897  {
898  case LISTEN_LISTEN:
900  break;
901  case LISTEN_UNLISTEN:
902  /* there is no Exec_UnlistenPreCommit() */
903  break;
904  case LISTEN_UNLISTEN_ALL:
905  /* there is no Exec_UnlistenAllPreCommit() */
906  break;
907  }
908  }
909  }
910 
911  /* Queue any pending notifies (must happen after the above) */
912  if (pendingNotifies)
913  {
914  ListCell *nextNotify;
915 
916  /*
917  * Make sure that we have an XID assigned to the current transaction.
918  * GetCurrentTransactionId is cheap if we already have an XID, but not
919  * so cheap if we don't, and we'd prefer not to do that work while
920  * holding NotifyQueueLock.
921  */
922  (void) GetCurrentTransactionId();
923 
924  /*
925  * Serialize writers by acquiring a special lock that we hold till
926  * after commit. This ensures that queue entries appear in commit
927  * order, and in particular that there are never uncommitted queue
928  * entries ahead of committed ones, so an uncommitted transaction
929  * can't block delivery of deliverable notifications.
930  *
931  * We use a heavyweight lock so that it'll automatically be released
932  * after either commit or abort. This also allows deadlocks to be
933  * detected, though really a deadlock shouldn't be possible here.
934  *
935  * The lock is on "database 0", which is pretty ugly but it doesn't
936  * seem worth inventing a special locktag category just for this.
937  * (Historical note: before PG 9.0, a similar lock on "database 0" was
938  * used by the flatfiles mechanism.)
939  */
940  LockSharedObject(DatabaseRelationId, InvalidOid, 0,
942 
943  /* Now push the notifications into the queue */
945 
946  nextNotify = list_head(pendingNotifies->events);
947  while (nextNotify != NULL)
948  {
949  /*
950  * Add the pending notifications to the queue. We acquire and
951  * release NotifyQueueLock once per page, which might be overkill
952  * but it does allow readers to get in while we're doing this.
953  *
954  * A full queue is very uncommon and should really not happen,
955  * given that we have so much space available in the SLRU pages.
956  * Nevertheless we need to deal with this possibility. Note that
957  * when we get here we are in the process of committing our
958  * transaction, but we have not yet committed to clog, so at this
959  * point in time we can still roll the transaction back.
960  */
961  LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
963  if (asyncQueueIsFull())
964  ereport(ERROR,
965  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
966  errmsg("too many notifications in the NOTIFY queue")));
967  nextNotify = asyncQueueAddEntries(nextNotify);
968  LWLockRelease(NotifyQueueLock);
969  }
970  }
971 }
972 
973 /*
974  * AtCommit_Notify
975  *
976  * This is called at transaction commit, after committing to clog.
977  *
978  * Update listenChannels and clear transaction-local state.
979  */
980 void
982 {
983  ListCell *p;
984 
985  /*
986  * Allow transactions that have not executed LISTEN/UNLISTEN/NOTIFY to
987  * return as soon as possible
988  */
989  if (!pendingActions && !pendingNotifies)
990  return;
991 
992  if (Trace_notify)
993  elog(DEBUG1, "AtCommit_Notify");
994 
995  /* Perform any pending listen/unlisten actions */
996  if (pendingActions != NULL)
997  {
998  foreach(p, pendingActions->actions)
999  {
1000  ListenAction *actrec = (ListenAction *) lfirst(p);
1001 
1002  switch (actrec->action)
1003  {
1004  case LISTEN_LISTEN:
1005  Exec_ListenCommit(actrec->channel);
1006  break;
1007  case LISTEN_UNLISTEN:
1008  Exec_UnlistenCommit(actrec->channel);
1009  break;
1010  case LISTEN_UNLISTEN_ALL:
1012  break;
1013  }
1014  }
1015  }
1016 
1017  /* If no longer listening to anything, get out of listener array */
1018  if (amRegisteredListener && listenChannels == NIL)
1020 
1021  /* And clean up */
1023 }
1024 
1025 /*
1026  * Exec_ListenPreCommit --- subroutine for PreCommit_Notify
1027  *
1028  * This function must make sure we are ready to catch any incoming messages.
1029  */
1030 static void
1032 {
1033  QueuePosition head;
1034  QueuePosition max;
1035  BackendId prevListener;
1036 
1037  /*
1038  * Nothing to do if we are already listening to something, nor if we
1039  * already ran this routine in this transaction.
1040  */
1042  return;
1043 
1044  if (Trace_notify)
1045  elog(DEBUG1, "Exec_ListenPreCommit(%d)", MyProcPid);
1046 
1047  /*
1048  * Before registering, make sure we will unlisten before dying. (Note:
1049  * this action does not get undone if we abort later.)
1050  */
1052  {
1054  unlistenExitRegistered = true;
1055  }
1056 
1057  /*
1058  * This is our first LISTEN, so establish our pointer.
1059  *
1060  * We set our pointer to the global tail pointer and then move it forward
1061  * over already-committed notifications. This ensures we cannot miss any
1062  * not-yet-committed notifications. We might get a few more but that
1063  * doesn't hurt.
1064  *
1065  * In some scenarios there might be a lot of committed notifications that
1066  * have not yet been pruned away (because some backend is being lazy about
1067  * reading them). To reduce our startup time, we can look at other
1068  * backends and adopt the maximum "pos" pointer of any backend that's in
1069  * our database; any notifications it's already advanced over are surely
1070  * committed and need not be re-examined by us. (We must consider only
1071  * backends connected to our DB, because others will not have bothered to
1072  * check committed-ness of notifications in our DB.)
1073  *
1074  * We need exclusive lock here so we can look at other backends' entries
1075  * and manipulate the list links.
1076  */
1077  LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
1078  head = QUEUE_HEAD;
1079  max = QUEUE_TAIL;
1080  prevListener = InvalidBackendId;
1082  {
1084  max = QUEUE_POS_MAX(max, QUEUE_BACKEND_POS(i));
1085  /* Also find last listening backend before this one */
1086  if (i < MyBackendId)
1087  prevListener = i;
1088  }
1092  /* Insert backend into list of listeners at correct position */
1093  if (prevListener > 0)
1094  {
1096  QUEUE_NEXT_LISTENER(prevListener) = MyBackendId;
1097  }
1098  else
1099  {
1102  }
1103  LWLockRelease(NotifyQueueLock);
1104 
1105  /* Now we are listed in the global array, so remember we're listening */
1106  amRegisteredListener = true;
1107 
1108  /*
1109  * Try to move our pointer forward as far as possible. This will skip
1110  * over already-committed notifications, which we want to do because they
1111  * might be quite stale. Note that we are not yet listening on anything,
1112  * so we won't deliver such notifications to our frontend. Also, although
1113  * our transaction might have executed NOTIFY, those message(s) aren't
1114  * queued yet so we won't skip them here.
1115  */
1116  if (!QUEUE_POS_EQUAL(max, head))
1118 }
1119 
1120 /*
1121  * Exec_ListenCommit --- subroutine for AtCommit_Notify
1122  *
1123  * Add the channel to the list of channels we are listening on.
1124  */
1125 static void
1126 Exec_ListenCommit(const char *channel)
1127 {
1128  MemoryContext oldcontext;
1129 
1130  /* Do nothing if we are already listening on this channel */
1131  if (IsListeningOn(channel))
1132  return;
1133 
1134  /*
1135  * Add the new channel name to listenChannels.
1136  *
1137  * XXX It is theoretically possible to get an out-of-memory failure here,
1138  * which would be bad because we already committed. For the moment it
1139  * doesn't seem worth trying to guard against that, but maybe improve this
1140  * later.
1141  */
1143  listenChannels = lappend(listenChannels, pstrdup(channel));
1144  MemoryContextSwitchTo(oldcontext);
1145 }
1146 
1147 /*
1148  * Exec_UnlistenCommit --- subroutine for AtCommit_Notify
1149  *
1150  * Remove the specified channel name from listenChannels.
1151  */
1152 static void
1153 Exec_UnlistenCommit(const char *channel)
1154 {
1155  ListCell *q;
1156 
1157  if (Trace_notify)
1158  elog(DEBUG1, "Exec_UnlistenCommit(%s,%d)", channel, MyProcPid);
1159 
1160  foreach(q, listenChannels)
1161  {
1162  char *lchan = (char *) lfirst(q);
1163 
1164  if (strcmp(lchan, channel) == 0)
1165  {
1166  listenChannels = foreach_delete_current(listenChannels, q);
1167  pfree(lchan);
1168  break;
1169  }
1170  }
1171 
1172  /*
1173  * We do not complain about unlistening something not being listened;
1174  * should we?
1175  */
1176 }
1177 
1178 /*
1179  * Exec_UnlistenAllCommit --- subroutine for AtCommit_Notify
1180  *
1181  * Unlisten on all channels for this backend.
1182  */
1183 static void
1185 {
1186  if (Trace_notify)
1187  elog(DEBUG1, "Exec_UnlistenAllCommit(%d)", MyProcPid);
1188 
1189  list_free_deep(listenChannels);
1190  listenChannels = NIL;
1191 }
1192 
1193 /*
1194  * ProcessCompletedNotifies --- send out signals and self-notifies
1195  *
1196  * This is called from postgres.c just before going idle at the completion
1197  * of a transaction. If we issued any notifications in the just-completed
1198  * transaction, send signals to other backends to process them, and also
1199  * process the queue ourselves to send messages to our own frontend.
1200  * Also, if we filled enough queue pages with new notifies, try to advance
1201  * the queue tail pointer.
1202  *
1203  * The reason that this is not done in AtCommit_Notify is that there is
1204  * a nonzero chance of errors here (for example, encoding conversion errors
1205  * while trying to format messages to our frontend). An error during
1206  * AtCommit_Notify would be a PANIC condition. The timing is also arranged
1207  * to ensure that a transaction's self-notifies are delivered to the frontend
1208  * before it gets the terminating ReadyForQuery message.
1209  *
1210  * Note that we send signals and process the queue even if the transaction
1211  * eventually aborted. This is because we need to clean out whatever got
1212  * added to the queue.
1213  *
1214  * NOTE: we are outside of any transaction here.
1215  */
1216 void
1218 {
1219  MemoryContext caller_context;
1220 
1221  /* Nothing to do if we didn't send any notifications */
1223  return;
1224 
1225  /*
1226  * We reset the flag immediately; otherwise, if any sort of error occurs
1227  * below, we'd be locked up in an infinite loop, because control will come
1228  * right back here after error cleanup.
1229  */
1231 
1232  /*
1233  * We must preserve the caller's memory context (probably MessageContext)
1234  * across the transaction we do here.
1235  */
1236  caller_context = CurrentMemoryContext;
1237 
1238  if (Trace_notify)
1239  elog(DEBUG1, "ProcessCompletedNotifies");
1240 
1241  /*
1242  * We must run asyncQueueReadAllNotifications inside a transaction, else
1243  * bad things happen if it gets an error.
1244  */
1246 
1247  /* Send signals to other backends */
1248  SignalBackends();
1249 
1250  if (listenChannels != NIL)
1251  {
1252  /* Read the queue ourselves, and send relevant stuff to the frontend */
1254  }
1255 
1256  /*
1257  * If it's time to try to advance the global tail pointer, do that.
1258  */
1260  {
1261  backendTryAdvanceTail = false;
1263  }
1264 
1266 
1267  MemoryContextSwitchTo(caller_context);
1268 
1269  /* We don't need pq_flush() here since postgres.c will do one shortly */
1270 }
1271 
1272 /*
1273  * Test whether we are actively listening on the given channel name.
1274  *
1275  * Note: this function is executed for every notification found in the queue.
1276  * Perhaps it is worth further optimization, eg convert the list to a sorted
1277  * array so we can binary-search it. In practice the list is likely to be
1278  * fairly short, though.
1279  */
1280 static bool
1281 IsListeningOn(const char *channel)
1282 {
1283  ListCell *p;
1284 
1285  foreach(p, listenChannels)
1286  {
1287  char *lchan = (char *) lfirst(p);
1288 
1289  if (strcmp(lchan, channel) == 0)
1290  return true;
1291  }
1292  return false;
1293 }
1294 
1295 /*
1296  * Remove our entry from the listeners array when we are no longer listening
1297  * on any channel. NB: must not fail if we're already not listening.
1298  */
1299 static void
1301 {
1302  Assert(listenChannels == NIL); /* else caller error */
1303 
1304  if (!amRegisteredListener) /* nothing to do */
1305  return;
1306 
1307  /*
1308  * Need exclusive lock here to manipulate list links.
1309  */
1310  LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
1311  /* Mark our entry as invalid */
1314  /* and remove it from the list */
1317  else
1318  {
1320  {
1322  {
1324  break;
1325  }
1326  }
1327  }
1329  LWLockRelease(NotifyQueueLock);
1330 
1331  /* mark ourselves as no longer listed in the global array */
1332  amRegisteredListener = false;
1333 }
1334 
1335 /*
1336  * Test whether there is room to insert more notification messages.
1337  *
1338  * Caller must hold at least shared NotifyQueueLock.
1339  */
1340 static bool
1342 {
1343  int nexthead;
1344  int boundary;
1345 
1346  /*
1347  * The queue is full if creating a new head page would create a page that
1348  * logically precedes the current global tail pointer, ie, the head
1349  * pointer would wrap around compared to the tail. We cannot create such
1350  * a head page for fear of confusing slru.c. For safety we round the tail
1351  * pointer back to a segment boundary (compare the truncation logic in
1352  * asyncQueueAdvanceTail).
1353  *
1354  * Note that this test is *not* dependent on how much space there is on
1355  * the current head page. This is necessary because asyncQueueAddEntries
1356  * might try to create the next head page in any case.
1357  */
1358  nexthead = QUEUE_POS_PAGE(QUEUE_HEAD) + 1;
1359  if (nexthead > QUEUE_MAX_PAGE)
1360  nexthead = 0; /* wrap around */
1361  boundary = QUEUE_POS_PAGE(QUEUE_TAIL);
1362  boundary -= boundary % SLRU_PAGES_PER_SEGMENT;
1363  return asyncQueuePagePrecedes(nexthead, boundary);
1364 }
1365 
1366 /*
1367  * Advance the QueuePosition to the next entry, assuming that the current
1368  * entry is of length entryLength. If we jump to a new page the function
1369  * returns true, else false.
1370  */
1371 static bool
1372 asyncQueueAdvance(volatile QueuePosition *position, int entryLength)
1373 {
1374  int pageno = QUEUE_POS_PAGE(*position);
1375  int offset = QUEUE_POS_OFFSET(*position);
1376  bool pageJump = false;
1377 
1378  /*
1379  * Move to the next writing position: First jump over what we have just
1380  * written or read.
1381  */
1382  offset += entryLength;
1383  Assert(offset <= QUEUE_PAGESIZE);
1384 
1385  /*
1386  * In a second step check if another entry can possibly be written to the
1387  * page. If so, stay here, we have reached the next position. If not, then
1388  * we need to move on to the next page.
1389  */
1391  {
1392  pageno++;
1393  if (pageno > QUEUE_MAX_PAGE)
1394  pageno = 0; /* wrap around */
1395  offset = 0;
1396  pageJump = true;
1397  }
1398 
1399  SET_QUEUE_POS(*position, pageno, offset);
1400  return pageJump;
1401 }
1402 
1403 /*
1404  * Fill the AsyncQueueEntry at *qe with an outbound notification message.
1405  */
1406 static void
1408 {
1409  size_t channellen = n->channel_len;
1410  size_t payloadlen = n->payload_len;
1411  int entryLength;
1412 
1413  Assert(channellen < NAMEDATALEN);
1414  Assert(payloadlen < NOTIFY_PAYLOAD_MAX_LENGTH);
1415 
1416  /* The terminators are already included in AsyncQueueEntryEmptySize */
1417  entryLength = AsyncQueueEntryEmptySize + payloadlen + channellen;
1418  entryLength = QUEUEALIGN(entryLength);
1419  qe->length = entryLength;
1420  qe->dboid = MyDatabaseId;
1421  qe->xid = GetCurrentTransactionId();
1422  qe->srcPid = MyProcPid;
1423  memcpy(qe->data, n->data, channellen + payloadlen + 2);
1424 }
1425 
1426 /*
1427  * Add pending notifications to the queue.
1428  *
1429  * We go page by page here, i.e. we stop once we have to go to a new page but
1430  * we will be called again and then fill that next page. If an entry does not
1431  * fit into the current page, we write a dummy entry with an InvalidOid as the
1432  * database OID in order to fill the page. So every page is always used up to
1433  * the last byte which simplifies reading the page later.
1434  *
1435  * We are passed the list cell (in pendingNotifies->events) containing the next
1436  * notification to write and return the first still-unwritten cell back.
1437  * Eventually we will return NULL indicating all is done.
1438  *
1439  * We are holding NotifyQueueLock already from the caller and grab
1440  * NotifySLRULock locally in this function.
1441  */
1442 static ListCell *
1444 {
1445  AsyncQueueEntry qe;
1446  QueuePosition queue_head;
1447  int pageno;
1448  int offset;
1449  int slotno;
1450 
1451  /* We hold both NotifyQueueLock and NotifySLRULock during this operation */
1452  LWLockAcquire(NotifySLRULock, LW_EXCLUSIVE);
1453 
1454  /*
1455  * We work with a local copy of QUEUE_HEAD, which we write back to shared
1456  * memory upon exiting. The reason for this is that if we have to advance
1457  * to a new page, SimpleLruZeroPage might fail (out of disk space, for
1458  * instance), and we must not advance QUEUE_HEAD if it does. (Otherwise,
1459  * subsequent insertions would try to put entries into a page that slru.c
1460  * thinks doesn't exist yet.) So, use a local position variable. Note
1461  * that if we do fail, any already-inserted queue entries are forgotten;
1462  * this is okay, since they'd be useless anyway after our transaction
1463  * rolls back.
1464  */
1465  queue_head = QUEUE_HEAD;
1466 
1467  /*
1468  * If this is the first write since the postmaster started, we need to
1469  * initialize the first page of the async SLRU. Otherwise, the current
1470  * page should be initialized already, so just fetch it.
1471  *
1472  * (We could also take the first path when the SLRU position has just
1473  * wrapped around, but re-zeroing the page is harmless in that case.)
1474  */
1475  pageno = QUEUE_POS_PAGE(queue_head);
1476  if (QUEUE_POS_IS_ZERO(queue_head))
1477  slotno = SimpleLruZeroPage(NotifyCtl, pageno);
1478  else
1479  slotno = SimpleLruReadPage(NotifyCtl, pageno, true,
1481 
1482  /* Note we mark the page dirty before writing in it */
1483  NotifyCtl->shared->page_dirty[slotno] = true;
1484 
1485  while (nextNotify != NULL)
1486  {
1487  Notification *n = (Notification *) lfirst(nextNotify);
1488 
1489  /* Construct a valid queue entry in local variable qe */
1491 
1492  offset = QUEUE_POS_OFFSET(queue_head);
1493 
1494  /* Check whether the entry really fits on the current page */
1495  if (offset + qe.length <= QUEUE_PAGESIZE)
1496  {
1497  /* OK, so advance nextNotify past this item */
1498  nextNotify = lnext(pendingNotifies->events, nextNotify);
1499  }
1500  else
1501  {
1502  /*
1503  * Write a dummy entry to fill up the page. Actually readers will
1504  * only check dboid and since it won't match any reader's database
1505  * OID, they will ignore this entry and move on.
1506  */
1507  qe.length = QUEUE_PAGESIZE - offset;
1508  qe.dboid = InvalidOid;
1509  qe.data[0] = '\0'; /* empty channel */
1510  qe.data[1] = '\0'; /* empty payload */
1511  }
1512 
1513  /* Now copy qe into the shared buffer page */
1514  memcpy(NotifyCtl->shared->page_buffer[slotno] + offset,
1515  &qe,
1516  qe.length);
1517 
1518  /* Advance queue_head appropriately, and detect if page is full */
1519  if (asyncQueueAdvance(&(queue_head), qe.length))
1520  {
1521  /*
1522  * Page is full, so we're done here, but first fill the next page
1523  * with zeroes. The reason to do this is to ensure that slru.c's
1524  * idea of the head page is always the same as ours, which avoids
1525  * boundary problems in SimpleLruTruncate. The test in
1526  * asyncQueueIsFull() ensured that there is room to create this
1527  * page without overrunning the queue.
1528  */
1529  slotno = SimpleLruZeroPage(NotifyCtl, QUEUE_POS_PAGE(queue_head));
1530 
1531  /*
1532  * If the new page address is a multiple of QUEUE_CLEANUP_DELAY,
1533  * set flag to remember that we should try to advance the tail
1534  * pointer (we don't want to actually do that right here).
1535  */
1536  if (QUEUE_POS_PAGE(queue_head) % QUEUE_CLEANUP_DELAY == 0)
1537  backendTryAdvanceTail = true;
1538 
1539  /* And exit the loop */
1540  break;
1541  }
1542  }
1543 
1544  /* Success, so update the global QUEUE_HEAD */
1545  QUEUE_HEAD = queue_head;
1546 
1547  LWLockRelease(NotifySLRULock);
1548 
1549  return nextNotify;
1550 }
1551 
1552 /*
1553  * SQL function to return the fraction of the notification queue currently
1554  * occupied.
1555  */
1556 Datum
1558 {
1559  double usage;
1560 
1561  /* Advance the queue tail so we don't report a too-large result */
1563 
1564  LWLockAcquire(NotifyQueueLock, LW_SHARED);
1565  usage = asyncQueueUsage();
1566  LWLockRelease(NotifyQueueLock);
1567 
1568  PG_RETURN_FLOAT8(usage);
1569 }
1570 
1571 /*
1572  * Return the fraction of the queue that is currently occupied.
1573  *
1574  * The caller must hold NotifyQueueLock in (at least) shared mode.
1575  */
1576 static double
1578 {
1579  int headPage = QUEUE_POS_PAGE(QUEUE_HEAD);
1580  int tailPage = QUEUE_POS_PAGE(QUEUE_TAIL);
1581  int occupied;
1582 
1583  occupied = headPage - tailPage;
1584 
1585  if (occupied == 0)
1586  return (double) 0; /* fast exit for common case */
1587 
1588  if (occupied < 0)
1589  {
1590  /* head has wrapped around, tail not yet */
1591  occupied += QUEUE_MAX_PAGE + 1;
1592  }
1593 
1594  return (double) occupied / (double) ((QUEUE_MAX_PAGE + 1) / 2);
1595 }
1596 
1597 /*
1598  * Check whether the queue is at least half full, and emit a warning if so.
1599  *
1600  * This is unlikely given the size of the queue, but possible.
1601  * The warnings show up at most once every QUEUE_FULL_WARN_INTERVAL.
1602  *
1603  * Caller must hold exclusive NotifyQueueLock.
1604  */
1605 static void
1607 {
1608  double fillDegree;
1609  TimestampTz t;
1610 
1611  fillDegree = asyncQueueUsage();
1612  if (fillDegree < 0.5)
1613  return;
1614 
1615  t = GetCurrentTimestamp();
1616 
1617  if (TimestampDifferenceExceeds(asyncQueueControl->lastQueueFillWarn,
1619  {
1620  QueuePosition min = QUEUE_HEAD;
1621  int32 minPid = InvalidPid;
1622 
1624  {
1626  min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
1627  if (QUEUE_POS_EQUAL(min, QUEUE_BACKEND_POS(i)))
1628  minPid = QUEUE_BACKEND_PID(i);
1629  }
1630 
1631  ereport(WARNING,
1632  (errmsg("NOTIFY queue is %.0f%% full", fillDegree * 100),
1633  (minPid != InvalidPid ?
1634  errdetail("The server process with PID %d is among those with the oldest transactions.", minPid)
1635  : 0),
1636  (minPid != InvalidPid ?
1637  errhint("The NOTIFY queue cannot be emptied until that process ends its current transaction.")
1638  : 0)));
1639 
1640  asyncQueueControl->lastQueueFillWarn = t;
1641  }
1642 }
1643 
1644 /*
1645  * Send signals to listening backends.
1646  *
1647  * We never signal our own process; that should be handled by our caller.
1648  *
1649  * Normally we signal only backends in our own database, since only those
1650  * backends could be interested in notifies we send. However, if there's
1651  * notify traffic in our database but no traffic in another database that
1652  * does have listener(s), those listeners will fall further and further
1653  * behind. Waken them anyway if they're far enough behind, so that they'll
1654  * advance their queue position pointers, allowing the global tail to advance.
1655  *
1656  * Since we know the BackendId and the Pid the signaling is quite cheap.
1657  */
1658 static void
1660 {
1661  int32 *pids;
1662  BackendId *ids;
1663  int count;
1664 
1665  /*
1666  * Identify backends that we need to signal. We don't want to send
1667  * signals while holding the NotifyQueueLock, so this loop just builds a
1668  * list of target PIDs.
1669  *
1670  * XXX in principle these pallocs could fail, which would be bad. Maybe
1671  * preallocate the arrays? But in practice this is only run in trivial
1672  * transactions, so there should surely be space available.
1673  */
1674  pids = (int32 *) palloc(MaxBackends * sizeof(int32));
1675  ids = (BackendId *) palloc(MaxBackends * sizeof(BackendId));
1676  count = 0;
1677 
1678  LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
1680  {
1681  int32 pid = QUEUE_BACKEND_PID(i);
1682  QueuePosition pos;
1683 
1684  Assert(pid != InvalidPid);
1685  if (pid == MyProcPid)
1686  continue; /* never signal self */
1687  pos = QUEUE_BACKEND_POS(i);
1689  {
1690  /*
1691  * Always signal listeners in our own database, unless they're
1692  * already caught up (unlikely, but possible).
1693  */
1694  if (QUEUE_POS_EQUAL(pos, QUEUE_HEAD))
1695  continue;
1696  }
1697  else
1698  {
1699  /*
1700  * Listeners in other databases should be signaled only if they
1701  * are far behind.
1702  */
1705  continue;
1706  }
1707  /* OK, need to signal this one */
1708  pids[count] = pid;
1709  ids[count] = i;
1710  count++;
1711  }
1712  LWLockRelease(NotifyQueueLock);
1713 
1714  /* Now send signals */
1715  for (int i = 0; i < count; i++)
1716  {
1717  int32 pid = pids[i];
1718 
1719  /*
1720  * Note: assuming things aren't broken, a signal failure here could
1721  * only occur if the target backend exited since we released
1722  * NotifyQueueLock; which is unlikely but certainly possible. So we
1723  * just log a low-level debug message if it happens.
1724  */
1725  if (SendProcSignal(pid, PROCSIG_NOTIFY_INTERRUPT, ids[i]) < 0)
1726  elog(DEBUG3, "could not signal backend with PID %d: %m", pid);
1727  }
1728 
1729  pfree(pids);
1730  pfree(ids);
1731 }
1732 
1733 /*
1734  * AtAbort_Notify
1735  *
1736  * This is called at transaction abort.
1737  *
1738  * Gets rid of pending actions and outbound notifies that we would have
1739  * executed if the transaction got committed.
1740  */
1741 void
1743 {
1744  /*
1745  * If we LISTEN but then roll back the transaction after PreCommit_Notify,
1746  * we have registered as a listener but have not made any entry in
1747  * listenChannels. In that case, deregister again.
1748  */
1749  if (amRegisteredListener && listenChannels == NIL)
1751 
1752  /* And clean up */
1754 }
1755 
1756 /*
1757  * AtSubCommit_Notify() --- Take care of subtransaction commit.
1758  *
1759  * Reassign all items in the pending lists to the parent transaction.
1760  */
1761 void
1763 {
1764  int my_level = GetCurrentTransactionNestLevel();
1765 
1766  /* If there are actions at our nesting level, we must reparent them. */
1767  if (pendingActions != NULL &&
1768  pendingActions->nestingLevel >= my_level)
1769  {
1770  if (pendingActions->upper == NULL ||
1771  pendingActions->upper->nestingLevel < my_level - 1)
1772  {
1773  /* nothing to merge; give the whole thing to the parent */
1774  --pendingActions->nestingLevel;
1775  }
1776  else
1777  {
1778  ActionList *childPendingActions = pendingActions;
1779 
1780  pendingActions = pendingActions->upper;
1781 
1782  /*
1783  * Mustn't try to eliminate duplicates here --- see queue_listen()
1784  */
1785  pendingActions->actions =
1786  list_concat(pendingActions->actions,
1787  childPendingActions->actions);
1788  pfree(childPendingActions);
1789  }
1790  }
1791 
1792  /* If there are notifies at our nesting level, we must reparent them. */
1793  if (pendingNotifies != NULL &&
1794  pendingNotifies->nestingLevel >= my_level)
1795  {
1796  Assert(pendingNotifies->nestingLevel == my_level);
1797 
1798  if (pendingNotifies->upper == NULL ||
1799  pendingNotifies->upper->nestingLevel < my_level - 1)
1800  {
1801  /* nothing to merge; give the whole thing to the parent */
1802  --pendingNotifies->nestingLevel;
1803  }
1804  else
1805  {
1806  /*
1807  * Formerly, we didn't bother to eliminate duplicates here, but
1808  * now we must, else we fall foul of "Assert(!found)", either here
1809  * or during a later attempt to build the parent-level hashtable.
1810  */
1811  NotificationList *childPendingNotifies = pendingNotifies;
1812  ListCell *l;
1813 
1814  pendingNotifies = pendingNotifies->upper;
1815  /* Insert all the subxact's events into parent, except for dups */
1816  foreach(l, childPendingNotifies->events)
1817  {
1818  Notification *childn = (Notification *) lfirst(l);
1819 
1820  if (!AsyncExistsPendingNotify(childn))
1821  AddEventToPendingNotifies(childn);
1822  }
1823  pfree(childPendingNotifies);
1824  }
1825  }
1826 }
1827 
1828 /*
1829  * AtSubAbort_Notify() --- Take care of subtransaction abort.
1830  */
1831 void
1833 {
1834  int my_level = GetCurrentTransactionNestLevel();
1835 
1836  /*
1837  * All we have to do is pop the stack --- the actions/notifies made in
1838  * this subxact are no longer interesting, and the space will be freed
1839  * when CurTransactionContext is recycled. We still have to free the
1840  * ActionList and NotificationList objects themselves, though, because
1841  * those are allocated in TopTransactionContext.
1842  *
1843  * Note that there might be no entries at all, or no entries for the
1844  * current subtransaction level, either because none were ever created, or
1845  * because we reentered this routine due to trouble during subxact abort.
1846  */
1847  while (pendingActions != NULL &&
1848  pendingActions->nestingLevel >= my_level)
1849  {
1850  ActionList *childPendingActions = pendingActions;
1851 
1852  pendingActions = pendingActions->upper;
1853  pfree(childPendingActions);
1854  }
1855 
1856  while (pendingNotifies != NULL &&
1857  pendingNotifies->nestingLevel >= my_level)
1858  {
1859  NotificationList *childPendingNotifies = pendingNotifies;
1860 
1861  pendingNotifies = pendingNotifies->upper;
1862  pfree(childPendingNotifies);
1863  }
1864 }
1865 
1866 /*
1867  * HandleNotifyInterrupt
1868  *
1869  * Signal handler portion of interrupt handling. Let the backend know
1870  * that there's a pending notify interrupt. If we're currently reading
1871  * from the client, this will interrupt the read and
1872  * ProcessClientReadInterrupt() will call ProcessNotifyInterrupt().
1873  */
1874 void
1876 {
1877  /*
1878  * Note: this is called by a SIGNAL HANDLER. You must be very wary what
1879  * you do here.
1880  */
1881 
1882  /* signal that work needs to be done */
1883  notifyInterruptPending = true;
1884 
1885  /* make sure the event is processed in due course */
1886  SetLatch(MyLatch);
1887 }
1888 
1889 /*
1890  * ProcessNotifyInterrupt
1891  *
1892  * This is called if we see notifyInterruptPending set, just before
1893  * transmitting ReadyForQuery at the end of a frontend command, and
1894  * also if a notify signal occurs while reading from the frontend.
1895  * HandleNotifyInterrupt() will cause the read to be interrupted
1896  * via the process's latch, and this routine will get called.
1897  * If we are truly idle (ie, *not* inside a transaction block),
1898  * process the incoming notifies.
1899  */
1900 void
1902 {
1904  return; /* not really idle */
1905 
1906  while (notifyInterruptPending)
1908 }
1909 
1910 
1911 /*
1912  * Read all pending notifications from the queue, and deliver appropriate
1913  * ones to my frontend. Stop when we reach queue head or an uncommitted
1914  * notification.
1915  */
1916 static void
1918 {
1919  volatile QueuePosition pos;
1920  QueuePosition head;
1921  Snapshot snapshot;
1922 
1923  /* page_buffer must be adequately aligned, so use a union */
1924  union
1925  {
1926  char buf[QUEUE_PAGESIZE];
1927  AsyncQueueEntry align;
1928  } page_buffer;
1929 
1930  /* Fetch current state */
1931  LWLockAcquire(NotifyQueueLock, LW_SHARED);
1932  /* Assert checks that we have a valid state entry */
1935  head = QUEUE_HEAD;
1936  LWLockRelease(NotifyQueueLock);
1937 
1938  if (QUEUE_POS_EQUAL(pos, head))
1939  {
1940  /* Nothing to do, we have read all notifications already. */
1941  return;
1942  }
1943 
1944  /*----------
1945  * Get snapshot we'll use to decide which xacts are still in progress.
1946  * This is trickier than it might seem, because of race conditions.
1947  * Consider the following example:
1948  *
1949  * Backend 1: Backend 2:
1950  *
1951  * transaction starts
1952  * UPDATE foo SET ...;
1953  * NOTIFY foo;
1954  * commit starts
1955  * queue the notify message
1956  * transaction starts
1957  * LISTEN foo; -- first LISTEN in session
1958  * SELECT * FROM foo WHERE ...;
1959  * commit to clog
1960  * commit starts
1961  * add backend 2 to array of listeners
1962  * advance to queue head (this code)
1963  * commit to clog
1964  *
1965  * Transaction 2's SELECT has not seen the UPDATE's effects, since that
1966  * wasn't committed yet. Ideally we'd ensure that client 2 would
1967  * eventually get transaction 1's notify message, but there's no way
1968  * to do that; until we're in the listener array, there's no guarantee
1969  * that the notify message doesn't get removed from the queue.
1970  *
1971  * Therefore the coding technique transaction 2 is using is unsafe:
1972  * applications must commit a LISTEN before inspecting database state,
1973  * if they want to ensure they will see notifications about subsequent
1974  * changes to that state.
1975  *
1976  * What we do guarantee is that we'll see all notifications from
1977  * transactions committing after the snapshot we take here.
1978  * Exec_ListenPreCommit has already added us to the listener array,
1979  * so no not-yet-committed messages can be removed from the queue
1980  * before we see them.
1981  *----------
1982  */
1983  snapshot = RegisterSnapshot(GetLatestSnapshot());
1984 
1985  /*
1986  * It is possible that we fail while trying to send a message to our
1987  * frontend (for example, because of encoding conversion failure). If
1988  * that happens it is critical that we not try to send the same message
1989  * over and over again. Therefore, we place a PG_TRY block here that will
1990  * forcibly advance our queue position before we lose control to an error.
1991  * (We could alternatively retake NotifyQueueLock and move the position
1992  * before handling each individual message, but that seems like too much
1993  * lock traffic.)
1994  */
1995  PG_TRY();
1996  {
1997  bool reachedStop;
1998 
1999  do
2000  {
2001  int curpage = QUEUE_POS_PAGE(pos);
2002  int curoffset = QUEUE_POS_OFFSET(pos);
2003  int slotno;
2004  int copysize;
2005 
2006  /*
2007  * We copy the data from SLRU into a local buffer, so as to avoid
2008  * holding the NotifySLRULock while we are examining the entries
2009  * and possibly transmitting them to our frontend. Copy only the
2010  * part of the page we will actually inspect.
2011  */
2012  slotno = SimpleLruReadPage_ReadOnly(NotifyCtl, curpage,
2014  if (curpage == QUEUE_POS_PAGE(head))
2015  {
2016  /* we only want to read as far as head */
2017  copysize = QUEUE_POS_OFFSET(head) - curoffset;
2018  if (copysize < 0)
2019  copysize = 0; /* just for safety */
2020  }
2021  else
2022  {
2023  /* fetch all the rest of the page */
2024  copysize = QUEUE_PAGESIZE - curoffset;
2025  }
2026  memcpy(page_buffer.buf + curoffset,
2027  NotifyCtl->shared->page_buffer[slotno] + curoffset,
2028  copysize);
2029  /* Release lock that we got from SimpleLruReadPage_ReadOnly() */
2030  LWLockRelease(NotifySLRULock);
2031 
2032  /*
2033  * Process messages up to the stop position, end of page, or an
2034  * uncommitted message.
2035  *
2036  * Our stop position is what we found to be the head's position
2037  * when we entered this function. It might have changed already.
2038  * But if it has, we will receive (or have already received and
2039  * queued) another signal and come here again.
2040  *
2041  * We are not holding NotifyQueueLock here! The queue can only
2042  * extend beyond the head pointer (see above) and we leave our
2043  * backend's pointer where it is so nobody will truncate or
2044  * rewrite pages under us. Especially we don't want to hold a lock
2045  * while sending the notifications to the frontend.
2046  */
2047  reachedStop = asyncQueueProcessPageEntries(&pos, head,
2048  page_buffer.buf,
2049  snapshot);
2050  } while (!reachedStop);
2051  }
2052  PG_FINALLY();
2053  {
2054  /* Update shared state */
2055  LWLockAcquire(NotifyQueueLock, LW_SHARED);
2057  LWLockRelease(NotifyQueueLock);
2058  }
2059  PG_END_TRY();
2060 
2061  /* Done with snapshot */
2062  UnregisterSnapshot(snapshot);
2063 }
2064 
2065 /*
2066  * Fetch notifications from the shared queue, beginning at position current,
2067  * and deliver relevant ones to my frontend.
2068  *
2069  * The current page must have been fetched into page_buffer from shared
2070  * memory. (We could access the page right in shared memory, but that
2071  * would imply holding the NotifySLRULock throughout this routine.)
2072  *
2073  * We stop if we reach the "stop" position, or reach a notification from an
2074  * uncommitted transaction, or reach the end of the page.
2075  *
2076  * The function returns true once we have reached the stop position or an
2077  * uncommitted notification, and false if we have finished with the page.
2078  * In other words: once it returns true there is no need to look further.
2079  * The QueuePosition *current is advanced past all processed messages.
2080  */
2081 static bool
2083  QueuePosition stop,
2084  char *page_buffer,
2085  Snapshot snapshot)
2086 {
2087  bool reachedStop = false;
2088  bool reachedEndOfPage;
2089  AsyncQueueEntry *qe;
2090 
2091  do
2092  {
2093  QueuePosition thisentry = *current;
2094 
2095  if (QUEUE_POS_EQUAL(thisentry, stop))
2096  break;
2097 
2098  qe = (AsyncQueueEntry *) (page_buffer + QUEUE_POS_OFFSET(thisentry));
2099 
2100  /*
2101  * Advance *current over this message, possibly to the next page. As
2102  * noted in the comments for asyncQueueReadAllNotifications, we must
2103  * do this before possibly failing while processing the message.
2104  */
2105  reachedEndOfPage = asyncQueueAdvance(current, qe->length);
2106 
2107  /* Ignore messages destined for other databases */
2108  if (qe->dboid == MyDatabaseId)
2109  {
2110  if (XidInMVCCSnapshot(qe->xid, snapshot))
2111  {
2112  /*
2113  * The source transaction is still in progress, so we can't
2114  * process this message yet. Break out of the loop, but first
2115  * back up *current so we will reprocess the message next
2116  * time. (Note: it is unlikely but not impossible for
2117  * TransactionIdDidCommit to fail, so we can't really avoid
2118  * this advance-then-back-up behavior when dealing with an
2119  * uncommitted message.)
2120  *
2121  * Note that we must test XidInMVCCSnapshot before we test
2122  * TransactionIdDidCommit, else we might return a message from
2123  * a transaction that is not yet visible to snapshots; compare
2124  * the comments at the head of heapam_visibility.c.
2125  *
2126  * Also, while our own xact won't be listed in the snapshot,
2127  * we need not check for TransactionIdIsCurrentTransactionId
2128  * because our transaction cannot (yet) have queued any
2129  * messages.
2130  */
2131  *current = thisentry;
2132  reachedStop = true;
2133  break;
2134  }
2135  else if (TransactionIdDidCommit(qe->xid))
2136  {
2137  /* qe->data is the null-terminated channel name */
2138  char *channel = qe->data;
2139 
2140  if (IsListeningOn(channel))
2141  {
2142  /* payload follows channel name */
2143  char *payload = qe->data + strlen(channel) + 1;
2144 
2145  NotifyMyFrontEnd(channel, payload, qe->srcPid);
2146  }
2147  }
2148  else
2149  {
2150  /*
2151  * The source transaction aborted or crashed, so we just
2152  * ignore its notifications.
2153  */
2154  }
2155  }
2156 
2157  /* Loop back if we're not at end of page */
2158  } while (!reachedEndOfPage);
2159 
2160  if (QUEUE_POS_EQUAL(*current, stop))
2161  reachedStop = true;
2162 
2163  return reachedStop;
2164 }
2165 
2166 /*
2167  * Advance the shared queue tail variable to the minimum of all the
2168  * per-backend tail pointers. Truncate pg_notify space if possible.
2169  */
2170 static void
2172 {
2173  QueuePosition min;
2174  int oldtailpage;
2175  int newtailpage;
2176  int boundary;
2177 
2178  /* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
2179  LWLockAcquire(NotifyQueueTailLock, LW_EXCLUSIVE);
2180 
2181  /* Compute the new tail. */
2182  LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
2183  min = QUEUE_HEAD;
2185  {
2187  min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
2188  }
2189  oldtailpage = QUEUE_POS_PAGE(QUEUE_TAIL);
2190  LWLockRelease(NotifyQueueLock);
2191 
2192  /*
2193  * We can truncate something if the global tail advanced across an SLRU
2194  * segment boundary.
2195  *
2196  * XXX it might be better to truncate only once every several segments, to
2197  * reduce the number of directory scans.
2198  */
2199  newtailpage = QUEUE_POS_PAGE(min);
2200  boundary = newtailpage - (newtailpage % SLRU_PAGES_PER_SEGMENT);
2201  if (asyncQueuePagePrecedes(oldtailpage, boundary))
2202  {
2203  /*
2204  * SimpleLruTruncate() will ask for NotifySLRULock but will also
2205  * release the lock again.
2206  */
2207  SimpleLruTruncate(NotifyCtl, newtailpage);
2208  }
2209 
2210  /*
2211  * Advertise the new tail. This changes asyncQueueIsFull()'s verdict for
2212  * the segment immediately prior to the new tail, allowing fresh data into
2213  * that segment.
2214  */
2215  LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
2216  QUEUE_TAIL = min;
2217  LWLockRelease(NotifyQueueLock);
2218 
2219  LWLockRelease(NotifyQueueTailLock);
2220 }
2221 
2222 /*
2223  * ProcessIncomingNotify
2224  *
2225  * Deal with arriving NOTIFYs from other backends as soon as it's safe to
2226  * do so. This used to be called from the PROCSIG_NOTIFY_INTERRUPT
2227  * signal handler, but isn't anymore.
2228  *
2229  * Scan the queue for arriving notifications and report them to my front
2230  * end.
2231  *
2232  * NOTE: since we are outside any transaction, we must create our own.
2233  */
2234 static void
2236 {
2237  /* We *must* reset the flag */
2238  notifyInterruptPending = false;
2239 
2240  /* Do nothing else if we aren't actively listening */
2241  if (listenChannels == NIL)
2242  return;
2243 
2244  if (Trace_notify)
2245  elog(DEBUG1, "ProcessIncomingNotify");
2246 
2247  set_ps_display("notify interrupt");
2248 
2249  /*
2250  * We must run asyncQueueReadAllNotifications inside a transaction, else
2251  * bad things happen if it gets an error.
2252  */
2254 
2256 
2258 
2259  /*
2260  * Must flush the notify messages to ensure frontend gets them promptly.
2261  */
2262  pq_flush();
2263 
2264  set_ps_display("idle");
2265 
2266  if (Trace_notify)
2267  elog(DEBUG1, "ProcessIncomingNotify: done");
2268 }
2269 
2270 /*
2271  * Send NOTIFY message to my front end.
2272  */
2273 void
2274 NotifyMyFrontEnd(const char *channel, const char *payload, int32 srcPid)
2275 {
2277  {
2279 
2280  pq_beginmessage(&buf, 'A');
2281  pq_sendint32(&buf, srcPid);
2282  pq_sendstring(&buf, channel);
2284  pq_sendstring(&buf, payload);
2285  pq_endmessage(&buf);
2286 
2287  /*
2288  * NOTE: we do not do pq_flush() here. For a self-notify, it will
2289  * happen at the end of the transaction, and for incoming notifies
2290  * ProcessIncomingNotify will do it after finding all the notifies.
2291  */
2292  }
2293  else
2294  elog(INFO, "NOTIFY for \"%s\" payload \"%s\"", channel, payload);
2295 }
2296 
2297 /* Does pendingNotifies include a match for the given event? */
2298 static bool
2300 {
2301  if (pendingNotifies == NULL)
2302  return false;
2303 
2304  if (pendingNotifies->hashtab != NULL)
2305  {
2306  /* Use the hash table to probe for a match */
2307  if (hash_search(pendingNotifies->hashtab,
2308  &n,
2309  HASH_FIND,
2310  NULL))
2311  return true;
2312  }
2313  else
2314  {
2315  /* Must scan the event list */
2316  ListCell *l;
2317 
2318  foreach(l, pendingNotifies->events)
2319  {
2320  Notification *oldn = (Notification *) lfirst(l);
2321 
2322  if (n->channel_len == oldn->channel_len &&
2323  n->payload_len == oldn->payload_len &&
2324  memcmp(n->data, oldn->data,
2325  n->channel_len + n->payload_len + 2) == 0)
2326  return true;
2327  }
2328  }
2329 
2330  return false;
2331 }
2332 
2333 /*
2334  * Add a notification event to a pre-existing pendingNotifies list.
2335  *
2336  * Because pendingNotifies->events is already nonempty, this works
2337  * correctly no matter what CurrentMemoryContext is.
2338  */
2339 static void
2341 {
2342  Assert(pendingNotifies->events != NIL);
2343 
2344  /* Create the hash table if it's time to */
2345  if (list_length(pendingNotifies->events) >= MIN_HASHABLE_NOTIFIES &&
2346  pendingNotifies->hashtab == NULL)
2347  {
2348  HASHCTL hash_ctl;
2349  ListCell *l;
2350 
2351  /* Create the hash table */
2352  MemSet(&hash_ctl, 0, sizeof(hash_ctl));
2353  hash_ctl.keysize = sizeof(Notification *);
2354  hash_ctl.entrysize = sizeof(NotificationHash);
2355  hash_ctl.hash = notification_hash;
2356  hash_ctl.match = notification_match;
2357  hash_ctl.hcxt = CurTransactionContext;
2358  pendingNotifies->hashtab =
2359  hash_create("Pending Notifies",
2360  256L,
2361  &hash_ctl,
2363 
2364  /* Insert all the already-existing events */
2365  foreach(l, pendingNotifies->events)
2366  {
2367  Notification *oldn = (Notification *) lfirst(l);
2368  NotificationHash *hentry;
2369  bool found;
2370 
2371  hentry = (NotificationHash *) hash_search(pendingNotifies->hashtab,
2372  &oldn,
2373  HASH_ENTER,
2374  &found);
2375  Assert(!found);
2376  hentry->event = oldn;
2377  }
2378  }
2379 
2380  /* Add new event to the list, in order */
2381  pendingNotifies->events = lappend(pendingNotifies->events, n);
2382 
2383  /* Add event to the hash table if needed */
2384  if (pendingNotifies->hashtab != NULL)
2385  {
2386  NotificationHash *hentry;
2387  bool found;
2388 
2389  hentry = (NotificationHash *) hash_search(pendingNotifies->hashtab,
2390  &n,
2391  HASH_ENTER,
2392  &found);
2393  Assert(!found);
2394  hentry->event = n;
2395  }
2396 }
2397 
2398 /*
2399  * notification_hash: hash function for notification hash table
2400  *
2401  * The hash "keys" are pointers to Notification structs.
2402  */
2403 static uint32
2404 notification_hash(const void *key, Size keysize)
2405 {
2406  const Notification *k = *(const Notification *const *) key;
2407 
2408  Assert(keysize == sizeof(Notification *));
2409  /* We don't bother to include the payload's trailing null in the hash */
2410  return DatumGetUInt32(hash_any((const unsigned char *) k->data,
2411  k->channel_len + k->payload_len + 1));
2412 }
2413 
2414 /*
2415  * notification_match: match function to use with notification_hash
2416  */
2417 static int
2418 notification_match(const void *key1, const void *key2, Size keysize)
2419 {
2420  const Notification *k1 = *(const Notification *const *) key1;
2421  const Notification *k2 = *(const Notification *const *) key2;
2422 
2423  Assert(keysize == sizeof(Notification *));
2424  if (k1->channel_len == k2->channel_len &&
2425  k1->payload_len == k2->payload_len &&
2426  memcmp(k1->data, k2->data,
2427  k1->channel_len + k1->payload_len + 2) == 0)
2428  return 0; /* equal */
2429  return 1; /* not equal */
2430 }
2431 
2432 /* Clear the pendingActions and pendingNotifies lists. */
2433 static void
2435 {
2436  /*
2437  * Everything's allocated in either TopTransactionContext or the context
2438  * for the subtransaction to which it corresponds. So, there's nothing to
2439  * do here except reset the pointers; the space will be reclaimed when the
2440  * contexts are deleted.
2441  */
2442  pendingActions = NULL;
2443  pendingNotifies = NULL;
2444 }
uint64 call_cntr
Definition: funcapi.h:65
#define DatumGetUInt32(X)
Definition: postgres.h:486
struct QueueBackendStatus QueueBackendStatus
#define NIL
Definition: pg_list.h:65
static void usage(void)
Definition: pg_standby.c:589
#define QUEUE_TAIL
Definition: async.c:288
BackendId firstListener
Definition: async.c:279
char data[NAMEDATALEN+NOTIFY_PAYLOAD_MAX_LENGTH]
Definition: async.c:176
bool XidInMVCCSnapshot(TransactionId xid, Snapshot snapshot)
Definition: snapmgr.c:2238
#define QUEUE_POS_IS_ZERO(x)
Definition: async.c:205
#define DEBUG1
Definition: elog.h:25
int MyProcPid
Definition: globals.c:40
int errhint(const char *fmt,...)
Definition: elog.c:1068
static void queue_listen(ListenActionKind action, const char *channel)
Definition: async.c:708
BackendId MyBackendId
Definition: globals.c:81
struct NotificationHash NotificationHash
List * events
Definition: async.c:397
#define pq_flush()
Definition: libpq.h:39
MemoryContext TopTransactionContext
Definition: mcxt.c:49
#define QUEUE_BACKEND_PID(i)
Definition: async.c:290
int page
Definition: async.c:189
#define HASH_CONTEXT
Definition: hsearch.h:91
#define HASH_ELEM
Definition: hsearch.h:85
uint32 TransactionId
Definition: c.h:521
Snapshot RegisterSnapshot(Snapshot snapshot)
Definition: snapmgr.c:810
static ListCell * lnext(const List *l, const ListCell *c)
Definition: pg_list.h:310
MemoryContext hcxt
Definition: hsearch.h:77
bool SlruScanDirCbDeleteAll(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.c:1426
#define DEBUG3
Definition: elog.h:23
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1578
static ActionList * pendingActions
Definition: async.c:357
void AsyncShmemInit(void)
Definition: async.c:518
int64 TimestampTz
Definition: timestamp.h:39
#define SRF_IS_FIRSTCALL()
Definition: funcapi.h:294
struct NotificationList NotificationList
char * pstrdup(const char *in)
Definition: mcxt.c:1187
void CommitTransactionCommand(void)
Definition: xact.c:2947
void SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
Definition: slru.c:1225
#define PG_RETURN_FLOAT8(x)
Definition: fmgr.h:365
char data[FLEXIBLE_ARRAY_MEMBER]
Definition: async.c:391
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
static void Exec_UnlistenAllCommit(void)
Definition: async.c:1184
int offset
Definition: async.c:190
Size entrysize
Definition: hsearch.h:72
MemoryContext CurTransactionContext
Definition: mcxt.c:50
#define FLEXIBLE_ARRAY_MEMBER
Definition: c.h:284
static List * listenChannels
Definition: async.c:325
List * list_concat(List *list1, const List *list2)
Definition: list.c:515
void AtPrepare_Notify(void)
Definition: async.c:854
int errcode(int sqlerrcode)
Definition: elog.c:610
Datum pg_notification_queue_usage(PG_FUNCTION_ARGS)
Definition: async.c:1557
bool IsTransactionOrTransactionBlock(void)
Definition: xact.c:4702
#define MemSet(start, val, len)
Definition: c.h:950
#define INFO
Definition: elog.h:33
static double asyncQueueUsage(void)
Definition: async.c:1577
void pq_sendstring(StringInfo buf, const char *str)
Definition: pqformat.c:197
void Async_Listen(const char *channel)
Definition: async.c:756
static void ClearPendingActionsAndNotifies(void)
Definition: async.c:2434
static void asyncQueueNotificationToEntry(Notification *n, AsyncQueueEntry *qe)
Definition: async.c:1407
Notification * event
Definition: async.c:406
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:919
static bool asyncQueueAdvance(volatile QueuePosition *position, int entryLength)
Definition: async.c:1372
#define QUEUE_HEAD
Definition: async.c:287
bool TransactionIdDidCommit(TransactionId transactionId)
Definition: transam.c:125
static void Exec_UnlistenCommit(const char *channel)
Definition: async.c:1153
unsigned int Oid
Definition: postgres_ext.h:31
#define PG_PROTOCOL_MAJOR(v)
Definition: pqcomm.h:104
void SetLatch(Latch *latch)
Definition: latch.c:505
bool TimestampDifferenceExceeds(TimestampTz start_time, TimestampTz stop_time, int msec)
Definition: timestamp.c:1677
Size SimpleLruShmemSize(int nslots, int nlsns)
Definition: slru.c:155
void list_free_deep(List *list)
Definition: list.c:1390
static bool IsListeningOn(const char *channel)
Definition: async.c:1281
#define SRF_PERCALL_SETUP()
Definition: funcapi.h:298
void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, LWLock *ctllock, const char *subdir, int tranche_id, SyncRequestHandler sync_handler)
Definition: slru.c:186
static bool unlistenExitRegistered
Definition: async.c:421
static bool asyncQueueIsFull(void)
Definition: async.c:1341
void pq_beginmessage(StringInfo buf, char msgtype)
Definition: pqformat.c:87
static bool AsyncExistsPendingNotify(Notification *n)
Definition: async.c:2299
static void asyncQueueFillWarning(void)
Definition: async.c:1606
signed int int32
Definition: c.h:363
int nestingLevel
Definition: async.c:396
#define PG_GETARG_TEXT_PP(n)
Definition: fmgr.h:308
static SlruCtlData NotifyCtlData
Definition: async.c:298
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1812
#define foreach_delete_current(lst, cell)
Definition: pg_list.h:357
#define list_make1(x1)
Definition: pg_list.h:206
QueuePosition pos
Definition: async.c:241
#define NAMEDATALEN
#define SRF_RETURN_NEXT(_funcctx, _result)
Definition: funcapi.h:300
void set_ps_display(const char *activity)
Definition: ps_status.c:349
static void pq_sendint32(StringInfo buf, uint32 i)
Definition: pqformat.h:145
#define NotifyCtl
Definition: async.c:300
static int asyncQueuePageDiff(int p, int q)
Definition: async.c:471
void PreCommit_Notify(void)
Definition: async.c:879
#define QUEUE_POS_OFFSET(x)
Definition: async.c:194
#define NUM_NOTIFY_BUFFERS
Definition: async.h:21
Definition: dynahash.c:218
int SendProcSignal(pid_t pid, ProcSignalReason reason, BackendId backendId)
Definition: procsignal.c:250
unsigned short uint16
Definition: c.h:374
void pfree(void *pointer)
Definition: mcxt.c:1057
static NotificationList * pendingNotifies
Definition: async.c:409
#define AsyncQueueEntryEmptySize
Definition: async.c:182
#define ERROR
Definition: elog.h:43
void PreventCommandDuringRecovery(const char *cmdname)
Definition: utility.c:444
void ProcessNotifyInterrupt(void)
Definition: async.c:1901
struct ActionList ActionList
#define QUEUE_FIRST_LISTENER
Definition: async.c:289
static void * list_nth(const List *list, int n)
Definition: pg_list.h:266
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:392
int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok, TransactionId xid)
Definition: slru.c:394
static void Async_UnlistenOnExit(int code, Datum arg)
Definition: async.c:841
BackendId nextListener
Definition: async.c:240
void AtSubCommit_Notify(void)
Definition: async.c:1762
static bool backendHasSentNotifications
Definition: async.c:427
int MaxBackends
Definition: globals.c:136
TransactionId GetCurrentTransactionId(void)
Definition: xact.c:438
static char * buf
Definition: pg_test_fsync.c:68
static bool backendTryAdvanceTail
Definition: async.c:430
#define QUEUE_PAGESIZE
Definition: async.c:301
#define SET_QUEUE_POS(x, y, z)
Definition: async.c:196
List * actions
Definition: async.c:353
static void AddEventToPendingNotifies(Notification *n)
Definition: async.c:2340
int errdetail(const char *fmt,...)
Definition: elog.c:954
void before_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:333
#define InvalidTransactionId
Definition: transam.h:31
static ListCell * list_head(const List *l)
Definition: pg_list.h:125
unsigned int uint32
Definition: c.h:375
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
static Datum hash_any(const unsigned char *k, int keylen)
Definition: hashfn.h:31
#define NOTIFY_PAYLOAD_MAX_LENGTH
Definition: async.c:156
static bool asyncQueuePagePrecedes(int p, int q)
Definition: async.c:492
struct NotificationList * upper
Definition: async.c:399
static AsyncQueueControl * asyncQueueControl
Definition: async.c:285
static void asyncQueueAdvanceTail(void)
Definition: async.c:2171
static void Exec_ListenCommit(const char *channel)
Definition: async.c:1126
#define IsParallelWorker()
Definition: parallel.h:61
static void SignalBackends(void)
Definition: async.c:1659
MemoryContext TopMemoryContext
Definition: mcxt.c:44
void UnregisterSnapshot(Snapshot snapshot)
Definition: snapmgr.c:852
#define QUEUE_FULL_WARN_INTERVAL
Definition: async.c:302
List * lappend(List *list, void *datum)
Definition: list.c:321
struct AsyncQueueControl AsyncQueueControl
#define WARNING
Definition: elog.h:40
#define MIN_HASHABLE_NOTIFIES
Definition: async.c:402
struct QueuePosition QueuePosition
Size mul_size(Size s1, Size s2)
Definition: shmem.c:515
#define PG_FINALLY()
Definition: elog.h:312
void AtSubAbort_Notify(void)
Definition: async.c:1832
#define InvalidBackendId
Definition: backendid.h:23
ListenActionKind
Definition: async.c:337
int nestingLevel
Definition: async.c:352
static void ProcessIncomingNotify(void)
Definition: async.c:2235
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:326
uintptr_t Datum
Definition: postgres.h:367
#define QUEUE_CLEANUP_DELAY
Definition: async.c:231
static void asyncQueueUnregister(void)
Definition: async.c:1300
static bool amRegisteredListener
Definition: async.c:424
void AtAbort_Notify(void)
Definition: async.c:1742
Size add_size(Size s1, Size s2)
Definition: shmem.c:498
int BackendId
Definition: backendid.h:21
#define QUEUEALIGN(len)
Definition: async.c:180
Oid MyDatabaseId
Definition: globals.c:85
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int pageno, TransactionId xid)
Definition: slru.c:494
Size keysize
Definition: hsearch.h:71
void LockSharedObject(Oid classid, Oid objid, uint16 objsubid, LOCKMODE lockmode)
Definition: lmgr.c:1017
HashCompareFunc match
Definition: hsearch.h:74
#define QUEUE_BACKEND_DBOID(i)
Definition: async.c:291
QueuePosition head
Definition: async.c:276
static bool asyncQueueProcessPageEntries(volatile QueuePosition *current, QueuePosition stop, char *page_buffer, Snapshot snapshot)
Definition: async.c:2082
#define InvalidOid
Definition: postgres_ext.h:36
static void asyncQueueReadAllNotifications(void)
Definition: async.c:1917
struct ActionList * upper
Definition: async.c:354
#define QUEUE_NEXT_LISTENER(i)
Definition: async.c:292
#define ereport(elevel,...)
Definition: elog.h:144
int GetCurrentTransactionNestLevel(void)
Definition: xact.c:857
#define PG_RETURN_VOID()
Definition: fmgr.h:348
Datum pg_notify(PG_FUNCTION_ARGS)
Definition: async.c:575
HTAB * hashtab
Definition: async.c:398
static int notification_match(const void *key1, const void *key2, Size keysize)
Definition: async.c:2418
#define PG_ARGISNULL(n)
Definition: fmgr.h:209
static ListCell * asyncQueueAddEntries(ListCell *nextNotify)
Definition: async.c:1443
#define Assert(condition)
Definition: c.h:746
void ProcessCompletedNotifies(void)
Definition: async.c:1217
#define lfirst(lc)
Definition: pg_list.h:169
bool SlruScanDirectory(SlruCtl ctl, SlruScanCallback callback, void *data)
Definition: slru.c:1449
static uint32 notification_hash(const void *key, Size keysize)
Definition: async.c:2404
uint16 channel_len
Definition: async.c:388
static void Exec_ListenPreCommit(void)
Definition: async.c:1031
void StartTransactionCommand(void)
Definition: xact.c:2846
uint16 payload_len
Definition: async.c:389
#define HASH_COMPARE
Definition: hsearch.h:88
struct AsyncQueueEntry AsyncQueueEntry
Size AsyncShmemSize(void)
Definition: async.c:501
size_t Size
Definition: c.h:474
void Async_UnlistenAll(void)
Definition: async.c:788
static int list_length(const List *l)
Definition: pg_list.h:149
struct Notification Notification
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1208
bool Trace_notify
Definition: async.c:433
Snapshot GetLatestSnapshot(void)
Definition: snapmgr.c:325
void Async_Notify(const char *channel, const char *payload)
Definition: async.c:609
Datum pg_listening_channels(PG_FUNCTION_ARGS)
Definition: async.c:808
TransactionId xid
Definition: async.c:174
char * text_to_cstring(const text *t)
Definition: varlena.c:221
#define AccessExclusiveLock
Definition: lockdefs.h:45
void AtCommit_Notify(void)
Definition: async.c:981
void HandleNotifyInterrupt(void)
Definition: async.c:1875
void * palloc(Size size)
Definition: mcxt.c:950
int errmsg(const char *fmt,...)
Definition: elog.c:821
void pq_endmessage(StringInfo buf)
Definition: pqformat.c:298
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:797
#define elog(elevel,...)
Definition: elog.h:214
char channel[FLEXIBLE_ARRAY_MEMBER]
Definition: async.c:347
int i
ListenActionKind action
Definition: async.c:346
#define CStringGetTextDatum(s)
Definition: builtins.h:86
int32 srcPid
Definition: async.c:175
void * arg
struct Latch * MyLatch
Definition: globals.c:54
volatile sig_atomic_t notifyInterruptPending
Definition: async.c:418
#define PG_FUNCTION_ARGS
Definition: fmgr.h:193
#define SLRU_PAGES_PER_SEGMENT
Definition: slru.h:34
TimestampTz lastQueueFillWarn
Definition: async.c:280
CommandDest whereToSendOutput
Definition: postgres.c:91
#define QUEUE_MAX_PAGE
Definition: async.c:318
#define PG_TRY()
Definition: elog.h:295
QueuePosition tail
Definition: async.c:277
#define QUEUE_BACKEND_POS(i)
Definition: async.c:293
Definition: pg_list.h:50
ProtocolVersion FrontendProtocol
Definition: globals.c:28
int SimpleLruZeroPage(SlruCtl ctl, int pageno)
Definition: slru.c:279
#define PG_END_TRY()
Definition: elog.h:320
#define QUEUE_POS_PAGE(x)
Definition: async.c:193
#define QUEUE_POS_EQUAL(x, y)
Definition: async.c:202
#define offsetof(type, field)
Definition: c.h:669
#define QUEUE_POS_MIN(x, y)
Definition: async.c:209
void Async_Unlisten(const char *channel)
Definition: async.c:770
void NotifyMyFrontEnd(const char *channel, const char *payload, int32 srcPid)
Definition: async.c:2274
HashValueFunc hash
Definition: hsearch.h:73
#define SRF_RETURN_DONE(_funcctx)
Definition: funcapi.h:318
#define HASH_FUNCTION
Definition: hsearch.h:87
#define InvalidPid
Definition: miscadmin.h:32
#define SRF_FIRSTCALL_INIT()
Definition: funcapi.h:296
#define QUEUE_POS_MAX(x, y)
Definition: async.c:215